development main phases:

• VHDL description of the coprocessor with Avalon MM interface
• Qsys construction of a Nios II system with coprocessor and performance counter
• system mapping to FPGA and compilation
• TCL script production for HAL software driver generation
• production of the software application for testing and performance measurement, in two versions:
  • sequential : blocking execution of the coprocessor computation
  • pipelined : nonblocking execution of the coprocessor computation
• compilation and execution of the application under the Monitor Program, for two variants of each version: one with defaut value of the optimization level, the other with level O3
• save of performance reports and project archiving

two VHDL sources implement the memory-mapped coprocessor:

• delay_collatz.vhd, modified version of the output by the fdlvhd translation of the Gezel source presented in the second lecture, according to the third lab experience
• delay_collatz_interface.vhd, which contains an instance of the computational component and accesses the following Avalon bus signals: clock, resetn, read, write, chipselect, waitrequest, writedata, readdata

both files are avaulable in the vhdl folder of the attached archive, as well as in the VHDL/code/e11 folder of the reserved lab area

• the folder also contains std_logic_arithext.vhd, which is needed to compile the computational component, delay_collatz_codesign.vhd, which will be explained next, and delay_collatz.diff, the latter for documentation only, to show the changes made to the translation produced by fdlvhd

consultation of the delay_collatz_interface.vhd source shows the relationships between the I/O signals of the computational component and the Avalon interface signals

folder codesign in the attached archive is preset to host the project development

• after having copied the *.vhd files from folder vhdl into it, the Qsys custom component construction goes much like in the tutorial seen in lab tutorial 10, with due differences for the present case

after creation of project delay_collatz_codesign, with top-level entity having the same name, the construction of the custom component delay_collatz_interface may proceed

• in particular, the Conduit Avalon interface is not needed by the present component, since it makes no use of peripherals outside the FPGA

the new component type definition is shown in the figure

the next step is the assignment of VHDL files that describe the component and their analysis, as shown in the figure

• N.B. for this project, it is not necessary to copy files for simulation

finally, the new component definition ends with the definition of its Avalon interfaces and placement of its signals under the appropriate interfaces, as shown in the figure

for the construction of the Nios II system shown in the previous figures it may be useful to consult the Qsys introduction tutorial

• with a few differences, such as: memory size is 128 KB in the present case, all base addresses are assigned by th system, etc.

the final steps to map the system to the FPGA are as follows:

in Qsys:

• save the system with name embedded_system by File > Save As...
• generate the VHDL code for it by Generate > Generate HDL...

exit Qsys, then in Quartus:

• assign the embedded_system.qip file (in embedded_system/synthesis) to the project
• import assignments from file DE1_SoC.qsf in folder de1soc of the attached archive
• File > Save Project
• compile delay_collatz_codesign.vhd

folder script in the attached archive contains two TCL scripts for the generation of the software driver in the BSP for the project

• the two scripts differ for a single command, present in one of them, that prescribes optimization level O3 rather than the default level O1

these two scripts are to be copied in folder codesign/ip/delay_collatz_avalon_interface

• in the same folder, respectively under HAL/inc and HAL/src, copy is to be made of the C sources delay_collatz_avalon_interface.h and delay_collatz_avalon_interface.c of the software driver, that are available in folder src of the attached archive

the TCL scripts were written by analogy with the TCL script for the software driver of the Performance Counter, available in the Quartus Prime Lite 16.1 distribution under path
$SOPC_KIT_NIOS2/../ip/altera/sopc_builder_ip/altera_avalon_performance_counter

• similarly, the C sources of the software driver were written by (more limited) analogy with the C sources of the software driver of the same IP Core, in folder HAL under the aforementioned path

the motivation for this, perhaps unorthodox, way of producing the software driver lies in the twofold fact that

• the Avalon interface of the custom component does not fit into any of the HAL generic device model classes defined in Chapter 7 of the Nios II Classic Software Developer's Handbook
• neither does the Performance Counter Unit IP Core fit therein ...

together with a somewhat reasonable level of operational analogy between the two components

• skimming through Chapter 7 of the handbook is recommended nonetheless, to get a better understanding of the software driver structure and contents

folder src in the attached archive contains the subject programs, which are to be copied in the provided folders for the creation of test and performance measurement projects under the Monitor Program, as follows:

• delay_collatz_sequential_timing.c in codesign/amp_s and in codesign/amp_s_o3
• delay_collatz_pipelined_timing.c in codesign/amp_p and in codesign/amp_p_o3

project creation parameters are summarized in the attached file MonitorNotes.txt

• the DE1-SoC needs to be powered-up and connected to the PC, to program the FPGA at the end of each project creation

main differences between the source of lab tutorial 09 and the present sequential version:

• #include and #define directives relating to the custom component
• replacement of the input from the switches device with a constant
• replacement of the body of function delay_collatz with two instructions from the software driver of the custom component

the pipelined version of the program exhibits much stronger differences with respect to the program of lab tutorial 09:

• the interaction with the custom hardware is made nonblocking by replacing the delay_collatz function call with an inlining of its body, yet where the software computation of the next trajectory start point is placed in between the two inlined instructions, respectively to start the hardware computation and to read its result

the synchronization mechanism is very simple, thanks to properties of the custom component and of the waitrequest signal of the Avalon MM protocol:

• for trajectories faster than the software computation, the custom component keeps the result in its internal register while waiting the read command
• for trajectories slower than the software computation, the read command is kept waiting by the Avalon interface by means of the waitrequest signal

compilation, loading on the FPGA and execution of program delay_collatz_sequential_timing.c, in the two projects codesign/amp_s and codesign/amp_s_o3, produces the Performance Counter Reports in the figure

• the remarkable reduction of the execution time of section delay_collatz in the second variant may be explained by the function inlining under compilation O3

a speed-up by an order of magnitude, w.r.t. the software computation in lab tutorial 09, results from the performance data in that case, with the same optimization levels

it is sensible to expect a further performance gain out of the nonblocking execution of the computation by the custom hardware

the comparison of the following Performance Counter Reports with the corresponding data for the implementation with all computation done in software, yields a 21x speed-up with default optimization O1 and a 16x speed-up with optimization O3; the corresponding speed-up values with blocking acceleration are 15x with O1 and 13x with O3

• N.B the speed-up is computed on the total time; section data are less significant with nonblocking acceleration because the execution threads of the two sections overlap in time

useful materials for the proposed lab experience:

• archive with source files for project reproduction
• Avalon® Interface Specifications, Ch. 1-3
  MNL-AVABUSREF, Intel Corp., 2017.05.08
• Making Qsys Components - For Quartus Prime 16.1
  Intel Corp. - FPGA University Program, November 2016
• Introduction to the Qsys System Integration Tool - For Quartus Prime 16.1
  Intel Corp. - FPGA University Program, November 2016
• Nios II Classic Software Developer's Handbook, Ch. 7
  NII5V2, Altera Corp., 2015.05.14
• Intel FPGA Monitor Program Tutorial for Nios II - For Quartus Prime 16.1
  Intel Corp. - FPGA University Program, November 2016